Open christianhuening opened 2 years ago
/assign ScheererJ
Hi Johannes, can you please check. @christianhuening can you please provide cluster details to Johannes.
@marwinski what details would you need beyond what's already stated above?
the cluster
@christianhuening Can you please try to switch to geneve
as tunnel protocol as see if that works for you? vxlan
as overlay seems to be broken in some configurations. At least, it was in the vsphere cluster I created.
spec:
...
networking:
type: cilium
providerConfig:
apiVersion: cilium.networking.extensions.gardener.cloud/v1alpha1
kind: NetworkConfig
tunnel: geneve
...
@christianhuening Does @ScheererJ's comment resolve this issue?
@briantopping unfortunately no, since the customer decided to switch over to Tanzu and hence we stopped using gardener+vsphere here.
How to categorize this issue?
/area networking /kind bug /platform vsphere
What happened: We installed a shooted Seed into vSphere using Calico as CNI solution. Then we created a Shoot based on that Seed using Cilium as CNI. The cluster got created and came up with one node and everything worked. When adding more nodes, specifically the
vsphere-csi-driver
pods stopped working because they couldn't communicate with the API server via its internal domain name (timeout).An interesting observation was that in vSphere the Cilium nodes apparently got two IPs, one from the node CIDR and one from the POD CIDR, which obviously is wrong.
With Calico the above works just fine.
Calico (node cidr: 10.0.0.0/16)
Cilium (node cidr: 10.10.0.0/16 & pod cidr: 10.80.0.0/12)
What you expected to happen: pods can communicate to api server properly and nodes dont get pod cidr IPs ;-)
How to reproduce it (as minimally and precisely as possible):
Environment:
kubectl version
): 1.23.9